How to rank in ChatGPT: LLM SEO for startups

Short answer: Think of ChatGPT search as search with a model in the middle. When it searches, it typically rewrites a user's question into short search queries, sends them to search providers and cites pages from the results. So allow OAI-SearchBot, the crawler behind ChatGPT search (GPTBot is a separate, training-only crawler). Get indexed by the search engines it uses. Publish pages that answer the search-style versions of your buyers' questions. Then get listed honestly on third-party pages that rank for those searches.

"How do I rank in ChatGPT?" is now the question founders ask right after "how do I rank on Google?" The internet is full of answers. Many are guesses presented as findings.

This post starts from what OpenAI, Perplexity, Anthropic and Microsoft have documented about how their AI search products find and cite sources. It adds a few third-party studies, labelled as such, and turns each mechanism into something a startup can do this week. LLM SEO and ChatGPT SEO are informal names for this work. The evidence behind each tactic is in our generative engine optimization guide.

What "ranking in ChatGPT" means

There's no list of ten blue links to climb. There's a contest inside each answer: whether ChatGPT searches at all, which pages it reads and which it cites. OpenAI says ChatGPT ranks search results using multiple factors and that placement isn't guaranteed. It doesn't say what the factors are. None of the vendors we read publishes ranking factors or weights for citations, so anyone claiming exact weights is guessing or selling.

What is documented is the mechanism: how the product searches, which crawler it uses and how it credits sources. That turns out to be enough to act on.

How ChatGPT search finds sources

OpenAI's help article on searching the web with ChatGPT describes the process.

• ChatGPT decides when to search. It may search automatically when a question would benefit from current information, or the user can pick Search. People who aren't signed in can use web search too, so test in a logged-out window.
• It rewrites the question. ChatGPT search "sometimes partners with other search providers", and when it does it "typically rewrites your query into one or more targeted queries" and sends those. OpenAI's example: a researcher asks about the latest on drugs that target CCR8 for cancer, and ChatGPT might first search "CCR8 immunotherapy drug development 2025".
• It may search again. After reviewing the first results, it "may send additional, more specific queries", such as "CHS-114 conference 2025".
• Memory can change the search. If memory is on, ChatGPT may use saved memories when rewriting a query, so two people asking the same thing can trigger different searches.
• It cites. Responses that use search may include citations, and a Sources button lists the cited pages and other relevant links. OpenAI warns that results and citations can be incomplete, outdated or wrong.

Notice the shape of the example. A conversational question can become a short, search-engine-style query. That's the query your page has to win.

OAI-SearchBot, GPTBot and ChatGPT-User

OpenAI's crawler documentation lists four user agents. Three matter for visibility, and they're easy to mix up. (The fourth, OAI-AdsBot, only checks pages submitted as ads on ChatGPT.)

Three more details from OpenAI's pages:

• The settings are independent. OpenAI's own example is allowing OAI-SearchBot to appear in search while disallowing GPTBot. If you allow both, it may use the results of one crawl for both. Allow roughly 24 hours for a robots.txt change to take effect for search. For ChatGPT-User, OpenAI says robots.txt rules may not apply, because a user starts the action, and points you to OAI-SearchBot for search opt-outs.
• Your host or CDN has to allow it too. The help article tells you to confirm your host or CDN allows OpenAI's published searchbot IP addresses. A robots.txt that welcomes the bot isn't enough if a firewall turns it away.
• Blocked pages can still show up as links. OpenAI's publisher FAQ says that if it learns a disallowed page's URL from a search provider or another page and has signals the page is relevant to a user's query, it may surface just the link and title in ChatGPT Atlas. Its fix is the noindex meta tag, which the crawler can only read if you let it crawl the page.

Sample robots.txt files for each choice are in our guide to robots.txt, llms.txt and AI crawlers.

What about Bing?

Founders ask this first, so here is what's documented.

OpenAI's consumer help article says ChatGPT search sometimes partners with other search providers, and links two of their privacy policies: Microsoft's and Shopify's. It doesn't publish a full provider list or say how much each contributes. The help page for Enterprise and Edu workspaces is more specific: ChatGPT "may share disassociated search queries with the Bing search engine" to return web results.

So Bing is documented as part of the pipeline for Enterprise and Edu workspaces (OpenAI says Healthcare workspaces don't use it), and Microsoft appears among the providers for the rest. Being indexed there is cheap insurance, not a guarantee. In Ahrefs' data from July 2025, about 8% of ChatGPT's cited URLs were in Bing's top 10 for the same prompt, and 6 to 8% were in Google's. Ahrefs' explanation is that assistants search several variations of the prompt, so how you rank for the user's exact words predicts little. It notes its results may have changed since.

Plenty of third-party articles say which index ChatGPT uses today. We found no OpenAI or Microsoft documentation for those claims, so we leave them out.

How Perplexity and Claude find sources

Perplexity. Its help center says it searches the internet in real time and answers with numbered citations that link to the original sources. Its crawler documentation describes two agents:

• PerplexityBot is "designed to surface and link websites in search results on Perplexity" and isn't used to crawl content for foundation models. Perplexity recommends allowing it in robots.txt and permitting its published IP ranges.
• Perplexity-User visits pages when users ask questions, and "generally ignores robots.txt rules" because a person started the request.

Perplexity hasn't published how it ranks sources. Many third-party articles claim to know. We haven't seen primary evidence for their specifics, so we won't repeat them.

Claude. Anthropic's help article says Claude searches when a topic benefits from current information and that answers using search include citations. Its crawler page names three bots: ClaudeBot for training, Claude-User for fetches when a user asks, and Claude-SearchBot, which it says improves search result quality. Blocking the last two may reduce your visibility in Claude. The help article names no provider for web results. It says only that image results are powered by Bing.

Where you can see citation data

Google and Microsoft now give site owners first-party data on how often AI features show their pages. OpenAI, Perplexity and Anthropic don't, as far as we found.

• Bing Webmaster Tools. The AI Performance report, launched in public preview in February 2026, shows how often your pages are cited in Microsoft Copilot and Bing's AI summaries, which pages, and the "grounding queries" the AI used when it retrieved them. Those are the closest thing to a keyword report for AI search. A June 2026 update added citation share and intent grouping. It covers Copilot, Bing's AI summaries and "select partner integrations". Microsoft doesn't say ChatGPT is one of them.
• Google Search Console. The Generative AI performance report shows impressions from AI Overviews and AI Mode by page. See AI Overviews and AI Mode for founders.
• ChatGPT. OpenAI's publisher FAQ says ChatGPT search adds utm_source=chatgpt.com to referral URLs, so you can filter your analytics by that source.

Microsoft's guidance on AI answers is worth reading too, because it's about what gets cited. It says there's no secret strategy, and that success starts with content that is fresh, authoritative, structured and clear.

What to do, mechanism by mechanism

1. Let the search crawlers in

If OAI-SearchBot is blocked, ChatGPT search won't show you. Perplexity recommends allowing PerplexityBot if you want to appear in its results. Check both, plus Claude-SearchBot and Bingbot, with our AI crawler checker.

Then check the network layer. OpenAI tells you to allow its published IP addresses, and Perplexity publishes a firewall setup guide. A firewall can refuse a bot before it ever reads robots.txt. If you use Cloudflare, its AI bot policies have separate Search, Agent and Training settings. Make sure Search isn't set to block.

2. Be indexed where they search

ChatGPT sends rewritten queries to search providers. OpenAI's pages point to Microsoft, and name Bing for Enterprise and Edu. So being indexed in Bing isn't something to skip.

• Verify your site in Bing Webmaster Tools. Bing lets you import verified sites from Google Search Console.
• Submit your sitemap there.
• Consider IndexNow, which notifies participating engines when pages change. Bing is one; Google isn't on its list of endpoints, so keep your sitemap honest for Google.

And keep doing normal SEO. Google's AI features rank on its core systems too. Our on-page SEO checklist is a good place to start.

3. Win the rewritten query

Since the assistant turns a chat into a search query, think about what that query looks like in your category. A buyer asking "what should I use to invoice clients as a freelancer in Germany?" probably triggers searches like "invoicing software freelancers Germany" or "best invoicing app EU VAT".

Pages that tend to match those searches:

• "Best X for Y" and category pages, where you're honestly one of several options.
• "X vs Y" comparisons with facts you've checked.
• "X alternatives" pages for the products people are leaving.
• Pricing and "how much does X cost" pages.
• Specific how-tos for problems your product solves.

Put the direct answer in the first paragraph, in the HTML, not behind a script or a tab. Microsoft's guidance warns that AI systems may not render content hidden in tabs or expandable menus. The query fan-out generator predicts the related searches for a topic. It's a prediction, not ChatGPT's real queries.

4. Be on the pages that already rank

Your own site is one candidate for a citation. The pages that already rank for "best X" are others, and you can't write them yourself. You can make sure you're on them, honestly:

• Relevant, reviewed directories.
• Launch platforms with permanent product pages. A free launch on LaunchRanked gives you one, with a followed link once approved.
• Comparison and review sites in your category.
• Communities where your category is discussed, as a real participant.

Keep the facts identical everywhere: name, one-line description, category, pricing model. When many independent pages agree on what you are, it's easier for any system to describe you correctly.

What not to do: Google's AI guide says chasing inauthentic "mentions" across the web does less than you'd expect. Fake reviews, planted forum posts and paid "AI citation" schemes are a liability, not a plan.

5. Write what can't be generated

A model already knows the generic version of every article. What it can't get elsewhere is your data, your experience and your honest comparisons. Google calls this non-commodity content, and says it will likely matter more in the long run than other suggestions. In the best-known benchmark study, adding citations, quotations from credible sources and statistics raised visibility in a test engine, while keyword stuffing didn't. A language model wrote those additions, so it's a lab result, covered in our GEO guide. Use real numbers only.

Measuring whether it's working

• Analytics: filter by utm_source=chatgpt.com, and check your referrers for perplexity.ai too.
• Bing Webmaster Tools: citations, cited pages and grounding queries for Copilot.
• Search Console: the Generative AI performance report for AI Overviews and AI Mode.
• A free check: the AI visibility checker asks ChatGPT, Claude, Perplexity and Gemini four questions a buyer might ask, with web search on, and shows which ones name and link you.
• Manual prompt tracking: a fixed list of buyer questions, checked on a schedule. We lay out the method in how to measure AI visibility without paid tools.

If you'd rather use software, we compare options on best AI visibility trackers. Our own Autopilot tracks 25 prompts across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and Google AI Mode every week, alongside the content work. The free trial covers Perplexity and Gemini.

If you use Claude, Cursor or Codex, LaunchRanked's MCP server lets the assistant check the technical side of this list for you: whether OAI-SearchBot and PerplexityBot may crawl your pages, whether the content is in the HTML, and your llms.txt. A coding agent can then fix what it finds in your code. Here's what that looked like on our own site.

The short version

1. Allow OAI-SearchBot, PerplexityBot, Claude-SearchBot and Bingbot, and check your firewall.
2. Get indexed in Bing as well as Google.
3. Publish pages that answer the search-style version of your buyers' questions.
4. Get listed, honestly, on third-party pages that rank for those searches.
5. Measure with UTM data, Bing's AI report and a fixed prompt list.

None of this is a hack. It's search, with a model in the middle. If you want the terms, our glossary covers ChatGPT search, LLM citations, generative engine optimization and AI share of voice.